MSc Thesis - UvA - Spatio-temporal Deep Learning approaches for hedonic price models and relation discovery

Master Thesis by David H. Chen

The Hedonic pricing method has been widely adopted for real estate appraisal and has been enhanced with machine learning (ML) and deep learning algorithms. Meanwhile, Hedonic models with time and space variant components have also been developed to capture the temporal and spatial auto-correlation in housing price. However, these ML-models are able to predict in training periods only.

This project aims to fill the gap. A novel hybrid method that utilises XG-Boost regressors and a Spatio-Temporal Neural Network (STNN) is introduced. Using a housing dataset of Amsterdam, we demonstrate how the model can be used to achieve out-of-training-period predictions. In addition, the model successfully unveils underlying spatial correlations amongst districts in the housing market of Amsterdam.